You still must have the block commit to the signature, or otherwise the signatures use in the network can be changed. (e.g. I could go add 200 GB to the size of the chain I give you by adding a megabyte of padding to the signatures in the first 200k blocks).  You also need all transaction relay to be changed to be in terms of a new ID which is the hash of the whole transaction, not the txid. (Otherwise I can take a transaction, corrupt the signature, and relay it to everyone fast to block the real transaction from the network.)

And when you've done that, tada you have segwit (specifically, you get the _exact_ construction that was used in elements alpha).  But a version of it with a ugly construction that requires every Bitcoin speaking program to upgrade _simultaneously_, and invalidates any pre-created nlocktimed transactions which _will_ confiscate some amount of users' Bitcoins (because some parties have been signing payments then destroying private keys as a time-lock-safe mechanism).

Exactly, thats why we were unable to go forward with the original segwit design that we used in elements alpha: It's nearly impossible to deploy. The improved design is just simply better and a big part of that-- though far from all of it-- is that it's actually deployable.